HAL Id: hal-00844408
https://hal.inria.fr/hal-00844408
Submitted on 15 Jul 2013
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Conditions of existence of oscillations for hybrid systems
Denis Efimov, Wilfrid Perruquetti, Anton Shiriaev
To cite this version:
Conditions of existence of oscillations for
hybrid systems
Denis Efimov∗,∗∗,∗∗∗∗Wilfrid Perruquetti∗,∗∗ Anton Shiriaev∗∗∗,∗∗∗∗
∗Non-A project at INRIA - LNE, Parc Scientifique de la Haute Borne,
40 avenue Halley, Bˆat.A Park Plaza, 59650 Villeneuve d’Ascq, France,{Denis.Efimov;Wilfrid.Perruquetti}@inria.fr
∗∗LAGIS (UMR-CNRS 8146), Ecole Centrale de Lille, BP 48, Cit´e
Scientifique, 59651 Villeneuve-d’Ascq, France
∗∗∗Department of Engineering Cybernetics, Norwegian University of
Science and Technology, NO-7491 Trondheim, Norway, e-mail:
Anton.Shiriaev@itk.ntnu.no
∗∗∗∗Department of Control Systems and Informatics, Saint Petersburg
State University of Information Technologies Mechanics and Optics (ITMO), Kronverkskiy av. 49, Saint Petersburg, 197101, Russia
Abstract: The paper extends the notion of oscillations in the sense of Yakubovich to hybrid dynamics. Several sufficient stability and instability conditions for a forward invariant set are presented. The consideration is motivated by analysis of a model of two-link compass-gait biped robot.
1. INTRODUCTION
Oscillations constitute one of the main operating modes for many systems in nature or in techniques Fransoise [2005], King [2009], Kurths et al. [2007], Burkin et al. [1996], Rogers [1999], Tass [2007]. In some cases it is required to maintain the oscillations, in other cases the oscillations have to be suppressed Chernousko and Frad-kov [2000], FradFrad-kov and Pogromsky [1998], Shiriaev and de Wit [2004]. In all cases the conditions of existence of sustained oscillations is of great importance since they al-low one to analyze/design a system with desired oscillating properties.
There are many stability theories and definitions of oscilla-tions Fradkov and Pogromsky [1998]. Among them in this work we choose one proposed by Prof. Yakubovich almost 40 years ago Yakubovich [1973]. This approach is rather generic, and it covers periodical and chaotic oscillations. Contrarily to a pure periodical case, when existence condi-tions are rather sophisticated Mallet-Paret and Sell [1996], the conditions of oscillations in the sense of Yakubovich (Y-oscillations) are simple. For Lurie systems, they are for-mulated in the frequency domain Yakubovich [1973, 1975], Yakubovich and Tomberg [1989], Burkin et al. [1996]. For a generic nonlinear system, the conditions of Y-oscillations are given using Lyapunov arguments Efimov and Fradkov [2009] or applying homogeneity framework Efimov and Perruquetti [2010]. The main goal of the present paper is to extend the notion of Y-oscillations to hybrid
dynam-? This work is partially supported by EU Interreg IV A 2 Mers Seas
Zee¨en Cross-border Cooperation Programme under SYSIASS project
06-020, and by Ministry of Higher Education and Research, Nord-Pas de Calais Regional Council and FEDER through the “Contrat de Projets Etat Region (CPER) CIA 2007-2013”.
ical systems. For this purpose several stability/instability Lyapunov conditions are formulated for hybrid dynamics. There exist many applications where the system has a hybrid dynamics (continuous and discontinuous) and it is oscillating. The most important one comes from robot locomotion Freidovich and Shiriaev [2010], Morris and Grizzle [2009], Spong et al. [2006]. The phenomenon has a hybrid nature, due to impacts occurring when a leg is hitting the ground, and the main operating mode is a periodical oscillation. The design/analysis of robot locomotion as a periodically oscillating system is rather sophisticated Freidovich and Shiriaev [2010], Morris and Grizzle [2009], Shiriaev et al. [2010], Shiriaev and de Wit [2004]. However, relaxing the periodicity requirement and considering Y-oscillations, it is possible to develop more constructive conditions for analysis and design of robot motion.
The problem is introduced and illustrated for a two-link compass-gait biped robot model in Section 2. Some preliminaries are given in Section 3. The main result on conditions of Y-oscillations in hybrid systems is presented in Section 4. An example is given in Section 5.
2. MOTIVATION
˙ x1= x2; p1x˙2− p2(c13x˙4+ s13x24) − p4sin(x1) = u; ˙ x3= x4; p3x˙4− p2(c13x˙2− s13x22) + p5sin(x3) = −u, x ∈ C; x+1 = x−3; x+3 = x−1; A x + 2 x+4 = B x − 2 x−4 , x ∈ D, (1)
where x = [x1x2x3x4]T ∈ R4 is the state vector of the
robot, u ∈ R is the control torque; s13 = sin(x1− x3),
c13= cos(x1− x3), A = p1− p2c − p 3− p2c− −p2c− p3 , B = p7c−− p6 −p6 −p6 0 , c− = cos(x−1 − x−3); C = R4 , D = {x ∈ R4 : cos(x 1+
ψ) = cos(x3 + ψ)} define the sets with continuous and
discrete dynamics respectively; pi, i = 1, 7 are the robot
parameters related with its physical counterparts pre-sented in Fig. 1 as p1 = (mH + m)l2+ ma2, p2 = mlb,
p3= mb2, p4 = (mHl + mb + ml)g, p5= mbg, p6 = mab,
p7= mHl2+ 2mal; ψ is the slope of the walking surface.
The standard abbreviations x− = x(t−) = lim
ε→0x(t − |ε|), x
+ = x(t+) = lim
ε→0x(t + |ε|)
are used to denote the values before and after jumps of a solution.
Fig. 1. Schematic of the compass-gait biped on a shallow slope
For ψ = 2.87π/180 and u = 0 the system has two (known) nontrivial periodic solutions Freidovich and Shiri-aev [2010]. If the value of ψ is changed or if it is a time-varying signal (that corresponds to a walking on an irregular surface), then the conditions of existence of periodical solutions is an open question. That is more, there is no practically applicable theoretical approach that can address this problem for a time-varying surface slope. The approaches, for a constant ψ, for detection of periodi-cal oscillating modes and their stabilization are largely re-ported in the literature (see Freidovich and Shiriaev [2010], Morris and Grizzle [2009], Shiriaev et al. [2010], Shiriaev and de Wit [2004] and references therein). This analysis and design methods are very complex and, frequently, only local. The problem complexity is originated by hybrid nature of the dynamics and by complicacy of the periodic motion. It is hard to analyze and stabilize such type of behavior. However, if we would skip the requirement that the motion has to be periodical, allowing other types of oscillating behavior (that is a natural relaxation for a time-varying ψ), then the oscillation existence conditions could
be more simple and global. We would like to demonstrate this developing Y-oscillation concept, which is introduced in the next section.
3. PRELIMINARY RESULTS
This section has three parts. The first one deals with the hybrid system formalism introduction following Cai et al. [2007, 2008], Ye et al. [1998]. The second part introduced (pre)asymptotic stability definition and its equivalent Lya-punov characterization from Cai et al. [2008]. The third part is devoted to presentation of Lyapunov sufficient conditions of instability for hybrid systems.
In this work, R denotes the real numbers, R+ the
non-negative real numbers, Z and Z+ are stated for integers
and nonnegative integers respectively. The symbol | · | denotes an absolute value for a real scalar or vector, |x|A= infy∈A|y − x| is the distance from a point x ∈ Rn
to a set A ⊂ Rn
. It is said that the function α : R+→ R+
belongs to class K if α(0) = 0 and it is strictly increasing, it is from class K∞ if α ∈ K and it is radially unbounded.
3.1 Hybrid systems
The considered hybrid system has the following form:
˙x ∈ F(x) x ∈ C;
x+∈ G(x) x ∈ D, (2) where x ∈ O ⊂ Rnis the state vector (it may contain
con-tinuously changing and discrete or logic-based variables); C ⊂ O and D ⊂ O are the sets where flow or jumps can occur respectively; the (set-valued) maps F : O ⇒ Rn and G : O ⇒ Rn determine the continuous and discrete
dynamics of the system (2). Following Cai et al. [2008] we impose the Standing Assumptions (SAs) on (2):
(SA1) the sets C ⊂ O and D ⊂ O are relatively closed in O (i.e. there exist some closed sets YC, YDsuch that
D = YD∩ O and C = YC∩ O);
(SA2) the map F : O ⇒ Rn is outer semicontinuous and locally bounded, F(x) is nonempty and convex for any x ∈ C;
(SA3) the map G : O ⇒ Rn is outer semicontinuous and
locally bounded, G(x) is nonempty subset of O for any x ∈ D.
These assumptions allow one to define solutions of the system (2) on hybrid time domains Cai et al. [2007, 2008]. A subset E ⊂ R+× Z+ is a compact hybrid time domain
if E =SJ −1
j=0([tj, tj+1], j) for some finite sequence of times
0 ≤ t0 ≤ . . . ≤ tJ. It is a hybrid time domain if for all
(T, J ) ∈ E, E ∩ ([0, T ] × {0, 1, . . . , J }) is a compact hybrid time domain. A hybrid arc is a function φ : E ⇒ O defined on a hybrid time domain E, and such that φ(·, j) is locally absolutely continuous for each j. A hybrid arc φ : domφ ⇒ O is a solution (trajectory) of (2) if φ(0, 0) ∈ C ∪ D and: (S1) ˙φ(t, j) ∈ F[φ(t, j)] for all j ∈ Z+ and almost all
t ∈ R+ such that (t, j) ∈ domφ and φ(t, j) ∈ C;
(S2) φ(t, j + 1) ∈ G[φ(t, j)] for all (t, j + 1) ∈ domφ and φ(t, j) ∈ D.
hybrid system is called nontrivial if domφ contains at least one point different from (0, 0); it is called maximal if its domain cannot be extended. A solution is called complete if its domain is unbounded (complete solutions are maximal), let S(x) denote the set of all maximal solutions for (2) starting from a point x ∈ O. The hybrid system (2) is said to be forward complete on O if for all x ∈ O, each φ ∈ S(x) is complete. The hybrid system (2) is said to be forward invariant on a set A ⊂ O if, for all x ∈ A, each φ ∈ S(x) is such that φ(t, j) ∈ A for all (t, j) ∈ domφ (this definition of forward invariance does not need existence of trajectories for all points of A). A point x0 is called an equilibrium of (2) if it is forward
invariant for the system.
If G(D) ∩ D 6= ∅, then multiple consequent switches or a Zeno-behavior are possible. In many cases (if D ⊂ C, for example) such a behavior does not correspond to physical nature of the considered system, but it is hard to avoid it in the formulation (2) since the sets C and D are closed. In addition, if D ⊂ C then on the intersection C ∩ D multiple solutions can be initiated, while frequently in practice only a jump is possible in D. Let us define the set of maximal solutions initiated from a point x ∈ O without multiple consequent switches with jumps in D as ˆS(x), i.e.
ˆ
S(x) = {φ ∈ S(x) : tj 6= tj+1, tj+1 = arg inft>tjφ(t, j) ∈
D, (tj+1, j + 1) ∈ domφ}. This set can be empty or,
if G(D) ∩ D = ∅ then ˆS(x) = S(x). Further, talking about a solution φ of the system (2) we will understand a maximal solution φ ∈ S(x) for some x ∈ O, except the restrictionφ ∈ ˆS(x) is explicitly stated.
3.2 Stability of hybrid systems
A continuous function ω : O → R+ is proper on O if
ω(xi) → +∞ when xi converges to a boundary of O
or |xi| → +∞. For a compact set A ⊂ O, the function
ω is a proper indicator for A on O if it is proper and A = {x ∈ O : ω(x) = 0}.
Definition 1. Cai et al. [2008] Let A ⊂ O be compact. (1) A is prestable for the system (2) if for each ε > 0,
there exists δ > 0 such that any solution φ with |φ(0, 0)|A ≤ δ satisfies |φ(t, j)|A ≤ ε for all (t, j) ∈
domφ.
(2) A is preattractive for the system (2) if there exists δ > 0 such that any solution φ with |φ(0, 0)|A ≤ δ
is bounded with respect to O, and if it is complete, then limt+j→+∞|φ(t, j)|A= 0.
(3) A is preasymptotically stable if it is prestable and preattractive.
(4) A is asymptotically stable if it is preasymptotically stable and there exists δ > 0 such that any maximal solution φ with |φ(0, 0)|A≤ δ is complete.
The set of all x ∈ O from which all solutions are bounded with respect to O and the complete ones converge to A is called the basin of preattraction and denoted as Oa
A. If
the conditions of Definition 1 are valid for any δ ∈ R+,
then the set A is globally preattractive/preasymptotically or asymptotically stable for (2).
Definition 2. A compact set A ⊂ O is called Non-Zeno (NZ) prestable (preattractive/preasymptotically sta-ble/asymptotically stable) for the system (2) if the
cor-responding properties of Definition 1 are satisfied for all φ ∈ ˆS only.
If for the system (2) the set A is (pre)asymptotically stable, then it is also NZ (pre)asymptotically stable, the converse is in general false.
Definition 3. Let X ⊂ O be open and ω : X → R+
be continuous. A function V : X → R+ is said to be a
smooth Lyapunov function for (X , F, G, C, D, ω) if there exist α1, α2∈ K∞ such that
α1(ω(x)) ≤ V (x) ≤ α2(ω(x)) ∀x ∈ X .
In this work we will use the following designations: Sup DV (x) = sup f ∈F(x) h∇V (x), f i , Inf DV (x) = inf f ∈F(x) h∇V (x), f i ; Sup V+(x) = sup g∈G(x)∩X V (g), Inf V+(x) = inf g∈G(x)∩X V (g).
This definition does not impose a restriction on derivative or increment of V since both stability and instability properties will be studied below.
Theorem 4. Cai et al. [2008] Let A be compact and X be open, A ⊂ X ⊂ O, and let ω be a proper indicator for A on X . The following statements are equivalent for (2): 1) There exists a smooth Lyapunov function V for (X , F, G, C, D, ω) such that
Sup DV (x) ≤ −V (x) ∀x ∈ X ∩ C; (3) Sup V+(x) ≤ e−1V (x) ∀x ∈ X ∩ D
and G(D ∩ X ) ⊂ X .
2) The set A is preasymptotically stable, its basin of preattraction Oa
Acontains X , and X is forward invariant
for (2).
Remark 5. Note that the sufficient part of this theorem holds if the condition (3) is replaced with the following one:
Sup DV (x) ≤ −ρ[ω(x)] ∀x ∈ X ∩ C; (4) Sup V+(x) ≤ λV (x) ∀x ∈ X ∩ D,
for some function ρ ∈ K and a constant λ ∈ [0, 1). Indeed, in this case V [φ(t, j)] ≤ σ1(V [φ(tj, j)])σ2(etj−t)
for all j ∈ Z+ and t ∈ [tj, tj+1] such that (t, j) ∈ domφ
and φ(t, j) ∈ C for some functions σ1, σ2 ∈ K; and
V [φ(t, j + 1)] ≤ λV [φ(t, j)] for all (t, j + 1) ∈ domφ and φ(t, j) ∈ D. Therefore, according to Definition 1 the system is prestable and preattractive with respect to the set A (A is preasymptotically stable).
Following Ye et al. [1998], a more simple for verification sufficient condition can be proposed for NZ preasymptot-ical stability of A based on the following assumption. Assumption 1. For all φ ∈ ˆS, sup(t,j)∈domφt = +∞ iff sup(t,j)∈domφj = +∞.
These conditions imply that the hybrid time domains of complete solutions in ˆS are unbounded with respect to both t and j. This assumption excludes from consideration a case with a finite number of jumps and the Zeno-behavior. The reasons for this condition introduction are the following. If the number of switches is finite for a complete solution φ ∈ ˆS, then the asymptotic transients are predefined by the continuous dynamics only. The Zeno-behavior is rather usual in hybrid models, but frequently it has no physical meaning, like in the compass-gait biped robot or the bouncing ball Cai et al. [2008], for instance. Theorem 7. Let A be compact and X be open, A ⊂ X ⊂ O, and let ω be a proper indicator for A on X . Let Assumption 1 be satisfied and there exist a differentiable Lyapunov function V for (X , F, G, C, D, ω) such that
Sup DV (x) ≤ 0 ∀x ∈ X ∩ C, (5) G(D ∩ X ) ⊂ X and for any solution φ ∈ ˆS for all (tj+1, j +
1) ∈ domφ such that φ(tj, j) ∈ X and φ(tj+1, j + 1) ∈ X
for some α ∈ K
V [φ(tj+1, j + 1)] − V [φ(tj, j)]
tj+1− tj
≤ −α(V [φ(tj+1, j + 1)]).(6)
Then the set A is NZ preasymptotically stable, its basin of preattraction OAa contains X , and X is forward invariant for (2).
All proofs are excluded due to space limitation.
Despite of the conditions of Theorem 7 include the inequal-ity (6) in the time domain for φ ∈ ˆS, in some cases it is more simple to apply Theorem 7 than Theorem 4 (that we will show later on the examples). Note that without the condition Sup V+(x) ≤ V (x) ∀x ∈ X ∩ D the function
V decreasing may be non-monotone, i.e. V [φ(tj+1, j +
1)] > V [φ(tj+1, j)] for some (tj+1, j + 1) ∈ domφ and
φ(tj+1, j) ∈ D. Therefore, the conditions of Theorem 7
admit a local increasing of V after jumps. 3.3 Instability of hybrid systems
This section contains a definition of instability (similarly to Forni and Teel [2010], Ye et al. [1998]) and the corre-sponding sufficient Lyapunov conditions.
Definition 8. Let U ⊂ O be compact.
(1) U is preunstable for the system (2) if it is not prestable in the sense of Definition 1, i.e. there exists ε > 0 such that for each δ > 0 there is a solution φ with |φ(0, 0)|U ≤ δ that satisfies |φ(t, j)|U > ε for
some (t, j) ∈ domφ.
(2) U is prerepulsive for the system (2) if there is some δ > 0 such that for any 0 < δ ≤ δ there exists a solution φ with |φ(0, 0)|U≤ δ for which there are some
(t0, j0) ∈ domφ such that the property |φ(t, j)|U > δ
holds for all (t, j) ∈ domφ with t ≥ t0, j ≥ j0.
(3) U is strictly prerepulsive for the system (2) if for some δ > 0 and any solution φ with |φ(0, 0)|U ≤ δ there
exist some (t0, j0) ∈ domφ such that the property |φ(t, j)|U > δ holds for all (t, j) ∈ domφ with t ≥ t0,
j ≥ j0.
As in the continuous case Hahn [1967], the hybrid system (2) can be preattractive with respect to a set A and preunstable with respect to this set simultaneously. If the set U is prerepulsive for the system (2), then it is preunstable, but this set still can be attractive for a some subset of initial states in a vicinity of U . If it is strictly prerepulsive, then all existent solutions exit a neighborhood of U . The domain of prerepulsion OrU contains x ∈ O such that φ(0, 0) = x and |φ(t, j)|U > δ
for all (t, j) ∈ domφwith t ≥ t0, j ≥ j0 ((t0, j0) ∈ domφ). Definition 9. A compact set U ⊂ O is called NZ preunsta-ble (prerepulsive/strictly prerepulsive) for the system (2) if the corresponding properties of Definition 8 are satisfied for φ ∈ ˆS only.
If U is NZ preunstable/prerepulsive for (2), then it is also preunstable/prerepulsive. If U is strictly prerepulsive, then it is also NZ strictly prerepulsive.
Theorem 10. Let U ⊂ X be compact and X ⊂ O be open, G(D ∩ X ) ⊂ X . Let a continuous function ω : X → R+ be
a proper indicator for U on X . Let also each φ ∈ S with φ(0, 0) ∈ X be nontrivial.
(1) The set U is strictly prerepulsive for the system (2), its domain of prerepulsion Or
U contains X , if there
exists a differentiable Lyapunov function U : X → R+
for (X , F, G, C, D, ω) such that for some function ρ ∈ K and a constant λ ∈ [1, +∞)
Inf DU (x) ≥ ρ[ω(x)] ∀x ∈ X ∩ C; (7) Inf U+(x) ≥ λU (x) ∀x ∈ X ∩ D.
(2) Let Assumption 1 be satisfied. The set U is NZ strictly prerepulsive for the system (2), its domain of prerepulsion Or
U contains X , if there exists a
differentiable Lyapunov function U : X → R+ for
(X , F, G, C, D, ω) such that for any solution φ ∈ ˆS for all (tj+1, j + 1) ∈ domφ such that φ(tj, j) ∈ X
and φ(tj+1, j + 1) ∈ X for some α ∈ K
U [φ(tj+1, j + 1)] − U [φ(tj, j)]
tj+1− tj
≥ α(U [φ(tj, j)]) (8)
and
Inf DU (x) ≥ 0 ∀x ∈ X ∩ C. (9) Note that the condition (8) is sufficient to prove that the set U is preunstable for the system (2). Contrarily the condition (7), the conditions (8), (9) do not restrict the Lyapunov function U behavior after switches (it may decrease on D).
4. OSCILLATION CONDITIONS
x1 ≤ x01, . . . , xn ≤ x0n implies that everywhere
ei-ther g(x1, . . . , xn) ≤ g(x01, . . . , x0n) or g(x1, . . . , xn) ≥
g(x01, . . . , x0n)).
Definition 11. Yakubovich [1973], Efimov and Fradkov [2009] A complete solution φ ∈ S(x) with x ∈ O of the system (2) is called [π−, π+]-oscillation with respect to the output ψ = η(x) (where η : O → R is a continuous monotone function) if
limt+j→+∞η[φ(t, j)] = π−; limt+j→+∞η[φ(t, j)] = π+;
−∞ < π− < π+< +∞.
A complete solution φ ∈ S(x) with x ∈ O of the system (2) is called oscillating, if there exist some output ψ and constants π−, π+ such that φ is [π−, π+]-oscillation with
respect to the output ψ. A forward complete system (2) is called oscillatory, if for almost all x ∈ O the solutions φ ∈ S(x) of the system are oscillating. The oscillatory system (2) is called uniformly oscillatory, if for almost all x ∈ O for corresponding solutions φ ∈ S(x) there exist the output ψ and the constants π−, π+ not depending on initial conditions.
In other words the solution φ is oscillating if the output ψ(t, j) = η[φ(t, j)] is asymptotically bounded and there is no single limit value of ψ(t, j) for t + j → +∞. The term “almost all solutions” is used to emphasize that generally the system (2) has a nonempty set of equilibrium points, thus there exists a set of initial conditions with zero measure such that the corresponding solutions are not oscillating. The notion of oscillations in the sense of Yakubovich is rather generic including periodical os-cillations (limit cycles), quasi-periodical, recurrent and chaotic trajectories. The oscillating trajectories could be repelling being oscillating. The trajectories also could be unbounded, it is required to find a function of the state vector η, that is bounded and admits certain requirements introduced in Definition 11.
Theorem 12. Let A ⊂ XA and U ⊂ XU ⊂ A be compact
and X = XA∪ XU ⊂ O be open. Let continuous functions
ωA: X → R+, ωU : X → R+ be proper indicators for A
and U on X respectively. Let G(D ∩ X ) ⊂ X , each φ ∈ S with φ(0, 0) ∈ X be nontrivial and the system (2) have two Lyapunov functions V and U such that for all x ∈ X :
υ1[ωA(x)] ≤ V (x) ≤ υ2[ωA(x)],
υ3[ωU(x)] ≤ U (x) ≤ υ4[ωU(x)],
υ1, υ2, υ3, υ4∈ K∞ and
1) the conditions (4) or (5), (6) with Assumption 1 are valid for the function V on XA;
2) the conditions (7) or (8), (9) with Assumption 1 are valid for the function U on XU;
3) Ω ∩ Ξ = ∅, where Ω = A ∩ XU and Ξ ⊂ Rn is a set with
zero Lebesgue measure containing all equilibriums of the system.
Then the system (2) has an oscillating solution.
Remark 13. If the conditions (4) and (7) are satisfied, then the system (2) is oscillating.
In the next section we will show an example of this theorem application.
5. EXAMPLE
Consider the following planar hybrid system: F(x) = (1 − |x|)x(1 − |x|)x1+ x2 2− x1 , G(x) = xx2 1 , (10) where x = [x1x2]T, O = R2, D = {x ∈ O : x1+ kx2= 0},
C = O and k ∈ R \ {−1, 1} is a parameter. Note that in this case D ⊂ C.
The system (10) has the single equilibrium at the origin for k 6= ±1 (Ξ = {0}). Next, it is easy to verify that the Lyapunov function W = |x|2for the continuous dynamics F has the derivative ˙W = 2(1 −√W )W for all x ∈ R2, then W (t, t0, W0) = [1 − (1 − W0−0.5)et0−t]−2 for any
initial condition W (t0) = W0, which according to Efimov
and Fradkov [2009], Efimov and Perruquetti [2010] implies existence of oscillating trajectories into the set W (x) = 1 (for ˙x = F(x)) and forward completeness of (10). To prove that for any initial condition x(t0, j) ∈ O, t0∈ R+, j ∈ Z+
for solutions of the system (10) always there exists a time instant t00≥ t0such that x(t00, j) ∈ D assume that it is not,
then from the expression of W (t, t0, W0) given above the
trajectories of the continuous dynamics converge to the set where W (x) = 1 and the continuous dynamics is reduced to a linear oscillator, thus t00 exists and Assumption 1 is satisfied.
For the chosen G we have that W (x+) = W (x) for all x ∈ D, therefore the value of W is not changing for the discrete dynamics. In addition, for k 6= ±1 the set D\{0} is not invariant for (10), therefore for the almost all instants of time the behavior of W is predefined by the continuous dynamics.
To apply Theorem 12 we can choose the origin as the set U , U (x) = W (x) and XU = {x ∈ O : |x| < 1}, then U (t, j)
is a strictly increasing function on XU and for any solution
φ ∈ ˆS of (10) with φ(0, 0) ∈ XUwe have U [φ(tj+1, j + 1)] −
U [φ(tj, j)] ≥ (tj+1− tj)2(1 −pU [φ(tj, j)])U [φ(tj, j)] for
all (tj, j) ∈ domφ while φ rests in XU, i.e. the conditions
of Theorem 10 are satisfied.
Let us select the set A = {x ∈ O : |x| ≤ 1}, then XA = O \ A. Using the function W we can prove that
this A is NZ preasymptotically stable for (10). Indeed, for any solution φ ∈ ˆS of (10) with φ(0, 0) ∈ XA we
have W [φ(tj+1, j + 1)] − W [φ(tj, j)] ≤ (tj+1− tj)2(1 −
pW [φ(tj+1, j + 1)])W [φ(tj+1, j +1)] for all (tj, j) ∈ domφ
while φ rests in XA. Select V (x) = max{0, W (x) − 1},
then outside of the set A we have V [φ(tj+1, j + 1)] −
V [φ(tj, j)] ≤ −(tj+1− tj)α{V [φ(tj+1, j + 1)]} for α(s) =
−(1 −√1 + s)(1 + s) ∈ K and all conditions of Theorem 7 are verified.
The set Ω = {x ∈ O : |x| = 1} does not contain the equilibrium of (10). Therefore, all conditions of Theorem 12 are satisfied and the system (10) has an oscillating solution.
signif-icantly (there exist two disjoint oscillating subsets in Ω for k = |1.5| and a discontinuous limit cycle for k = |0.5|).
2 − −1 0 1 2 2 − 1 − 0 1 2 2 − −1 0 1 2 2 − 1 − 0 1 2 2 − −1 0 1 2 2 − 1 − 0 1 2 2 − −1 0 1 2 2 − 1 − 0 1 2 k=0.5 k=-0.5 k=1.5 k=-1.5
Fig. 2. The hybrid system (10) trajectories for k ∈ {−0.5, 0.5, −1.5, 1.5}
6. CONCLUSIONS
The paper develops the framework of oscillations in the sense of Yakubovich to the class of hybrid systems (having continuous and discrete dynamics). Some sufficient Lya-punov conditions for stability/instability of hybrid sys-tems are presented. The utility of the proposed theory is demonstrated on an example. Extensions to different robotic systems are planned in future works.
REFERENCES
I.M. Burkin, G.A. Leonov, and A.I. Shepelyavyi. Fre-quency Methods in Oscillation Theory. Mathematics and Its Applications. Kluwer, Dordrecht, 1996. in Rus-sian: 1992.
C. Cai, A.R. Teel, and R. Goebel. Smooth lyapunov func-tions for hybrid systems part i: Existence is equivalent to robustness. IEEE Trans. Autom. Control, 52(7):1264– 1277, 2007.
C. Cai, A.R. Teel, and R. Goebel. Smooth lyapunov functions for hybrid systems part ii: (pre)asymptotically stable compact sets. IEEE Trans. Aut. Contr., 53(3): 734–748, 2008.
F.L. Chernousko and A.L. Fradkov, editors. Control of Oscillations and Chaos: 2000 2nd International Con-ference Proceedings. IEEE, November 2000.
D. Efimov and A.L. Fradkov. Oscillatority of nonlinear systems with static feedback. SIAM J. Control Opti-mization, 48(2):618–640, 2009.
D. Efimov and W. Perruquetti. Oscillations conditions in homogenous systems. In Proc. IFAC NOLCOS Symp., pages 1379–1384, Bologna, 2010.
Fulvio Forni and Andrew R. Teel. Instability and over-shoots of solutions for a class of homogeneous hybrid systems by lyapunov-like analysis. In Proc. 49th IEEE
Conference on Decision and Control (CDC), pages 2390–2395, Atlanta, 2010.
A.L. Fradkov and A.Yu. Pogromsky. Introduction to oscillations and chaos. World Scientific, Singapore, 1998.
Jean-Pierre Fransoise. Oscillations En Biologie: Analyse Qualitative Et Modeles. Springer-Verlag, August 2005. L. Freidovich and A.S. Shiriaev. Transverse linearization
for an underactuated compass-like biped robot and analysis of the closed-loop system. In Proc. the 8th IFAC Symposium on Nonlinear Control Systems (NOLCOS), Bologna, August 2010. IFAC.
W. Hahn. Stability of Motion. Stability of Motion, NY, 1967.
George C. King. Vibrations and Waves. John Wiley & Sons Inc., August 2009.
Jurgen Kurths, Grigory V. Osipov, and Changsong Zhou. Synchronization in Oscillatory Networks. Springer Se-ries in Synergetics. Springer-Verlag, September 2007. J. Mallet-Paret and G.R. Sell. The poincar-bendixson
theorem for monotone cyclic feedback systems with delay. J. Differential Equations, 125:441–489, 1996. B. Morris and J.W. Grizzle. Hybrid invariant manifolds in
systems with impulse effects with application to periodic locomotion in bipedal robots. IEEE Transaction on Automatic Control, 54(8):1751–1764, 2009.
Graham Rogers. Power System Oscillations. The Kluwer International Series in Engineering and Computer Sci-ence. Kluwer Academic Pub, 1999.
A.S. Shiriaev and C. Canudas de Wit. Virtual constraints: a tool for orbital stabilization of nonlinear systems theory. In Proc. 6th IFAC Symposium NOLCOS 2004, pages 1355–1360, Stuttgart, 2004. IFAC.
A.S. Shiriaev, L. Freidovich, and S. Gusev. Comput-ing a transverse linearization for mechanical systems with two and more underactuated degrees of freedom. IEEE Transactions on Automatic Control, 55(4):893– 906, 2010.
M. W. Spong, S. Hutchinson, and M. Vidyasagar. Robot Modeling and Control. John Wiley & Sons, New York, 2006.
Peter A. Tass. Phase Resetting in Medicine and Biology: Stochastic Modelling and Data Analysis. Springer Series in Synergetics. Springer-Verlag, January 2007.
V.A. Yakubovich. Frequency oscillations conditions in nonlinear systems with stationary single nonlinearity. Siberian Math. J., 14(2), 1973.
V.A. Yakubovich. Oscillations in systems with discon-tinuous and hysteresis nonlinearities. Automation and Remote Control, 12, 1975.
V.A. Yakubovich and E.A. Tomberg. Conditions for self-induced oscillations in nonlinear systems. Siberian Math. J., 30:641–653, 1989.
